UltraHi-PrNet: An Ultra-High Precision Deep Learning Network for Dense Multi-Scale Target Detection in SAR Images

Round 1

Reviewer 1 Report

This paper designed an ultra-high precision deep learning 104

Network (Ultrahi-Prnet) that can detect dense objects of different scales in SAR images. This is a very meaningful work for SAR image detection.

I think it can be published, but I have a question about whether the robustness performance of the model has been tested on other remote sensing datasets.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The manuscript "UltraHi-PrNet: An Ultra-High Precision Deep Learning Network for Dense Multi-Scale Target Detection in SAR Images" proposes the utilization of feature extraction layer suitable for targets of different scales, followed by a traditional classification layer,  and performance is evaluated against a combination of multiple SAR datasets.

The text is well written and easy to follow. A more explicit definition of "ultra-large" and "ultra-small" targets could help the reader with better context. It is also useful to summarise the images used regarding their resolution and calibration. Table-1 (page 14) and Table-2 (page 14) are easier to read if referenced to "The original method", as used in Table-3 (page 16)

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Point 1: The UltraHi-PrNet architecture proposed in the paper is based on the FPN architecture, and the effect is achieved in Faster R-CNN through the RestNet-101 backbone network, but the Faster R-CNN network is not mentioned in the abstract or elsewhere in the paper. If the detection is implemented based on Faster R-CNN, there is a necessary addition of references and descriptions in the paper. Is the backbone network missing from the overall algorithm framework architecture given in Figure 2? The role of the backbone network is not reflected in Figure 2.

 

Point 2: What are the classification loss function and regression loss function mentioned in Line355 used?

Point 3: P, R, AP, and mAP are proposed in the experimental metrics, but why are only the results of R and mAP given in the table? From the experimental data set of the paper, the data in this paper include airport and ship data. It is hoped that more detailed experimental result data will be available to list the detection results of different targets. Tables 1-5 need to list more detailed detection results.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

The authors responded to my suggestions and comments appropriately. Therefore, I have no further suggestions or comments.